ai2-kit feat nmrnet predict --help
src | Source code of original paper
To run the NMRNet tool, you need to install extra packages besides ai2-kit
since the extra dependencies are too large to be included.
You can install the extra packages by running the following command:
# Install rdkit
pip install rdkit
# Install pytorch
# You can find your CUDA version from
# https://pytorch.org/get-started/previous-versions/
pip install torch==2.1.2 torchvision==0.16.2 torchaudio==2.1.2 --index-url https://download.pytorch.org/whl/cu118
# Install unicore
# You should download the version that matches your CUDA and Python from:
# https://github.com/dptech-corp/Uni-Core/releases
wget https://github.com/dptech-corp/Uni-Core/releases/download/0.0.3/unicore-0.0.1+cu118torch2.0.0-cp39-cp39-linux_x86_64.whl
pip install unicore-0.0.1+cu118torch2.0.0-cp39-cp39-linux_x86_64.whl
# Downgrade numpy to 1.x to workaround the compatibility issue of uni-core
pip install numpy==1.24.3
You can find the parameters of the predict
command by running the following command:
ai2-kit feat nmrnet predict --help
To run the tool, you need to prepare extra files for the following parameters:
model_path
: NMRNet model file.dict_path
: NMRNet dictionary file.scaler_path
: Scaler file, for exampletarget_scaler.ss
.
An example of running the tool is as follows:
ai2-kit feat nmrnet predict \
--model_path ./weight/cv_seed_42_fold_0/checkpoint_best.pt \
--dict_path ./weight/oc_limit_dict.txt \
--scaler_path ./weight/target_scaler.ss \
--nmr_type solid --selected_atom H \
--data_file tmp/nmr-demo.xyz
The above command will predict the NMR of the 'selected_atom' atom in the 'data_file' file. The prediction result will be printed to stdout, and the chemical shifts will be displayed in the same order as the corresponding atoms in the xyz file. And then you can follow it with a demo result like this:
[25.572277 29.441292 29.615028 29.896814 29.590061 30.400713 29.576017
15.920061 29.084885 29.069242 28.565197 28.77764 28.510954 28.849863
14.738946 24.312784]
- Xu F, Guo W, Wang F, et al. Towards a Unified Benchmark and Framework for Deep Learning-Based Prediction of Nuclear Magnetic Resonance Chemical Shifts[J]. arXiv preprint arXiv:2408.15681, 2024.
Note: Our work has recently been accepted by Nature Computational Science, and we will update the reference once it is published.